Scholarly Works and Research
Millions of books were bought, then cut apart
In a warehouse, a machine takes the spine off a book, the loose pages run through a scanner, and the paper goes to recycling. The American company Anthropic did this to millions of print books it had purchased, to train its AI model. Court documents show it hired Tom Turvey, formerly of Google Books, in February 2024 and tasked him with obtaining “all the books in the world.” [1]
A judge ruled that destroying the book made it legal
In June 2025 a US district judge, William Alsup, found the practice lawful. Once you buy a book it is yours, and because the paper original was destroyed, the scan didn’t add a copy, it replaced one.
The uncomfortable part sits right there. Had the company kept the books on a shelf, it would have held two copies and broken the law. Shredding them is what kept it legal. [2]
Seven million pirated books cost $1.5 billion
Before it bought anything, the same company had already pulled more than seven million books off piracy sites. [3] The judge drew a hard line there, and it ended in a $1.5 billion settlement: roughly $3,000 per book across more than 400,000 works. [4]
Anything printed before 2022 is now the prize
So much writing online is machine-produced that models risk feeding on their own output, and quality degrades. Books printed before 2022 carry none of that contamination, which is what makes old stock valuable.
404 Media reported in July 2026 that one company was offering AI firms bulk orders of 1,000 to a million books. That company later removed the page and said no such service ever launched. In the same report, a bookseller said the orders clear stock that would never sell, but that uncommon books are being pulped. [5]
Who else is doing this, beyond Anthropic, is not established. Snopes rated the wider claim mostly true, with parts undetermined. [6]
64% of Americans still read print
Pew Research Center asked US adults in October 2025 what they had read in the past year. 64% had read a print book, 31% an e-book, 26% an audiobook. Print is the only format a majority still uses. [7]
A book on your shelf depends on no server
Digitisation was sold to us as preservation. What happened instead was that money was spent to buy books and turn them into pulp. A digital copy survives only as long as someone keeps paying for the server; it can be edited or quietly removed, and you will not be told.
Buy the hard copy, and keep it. Especially the ones nobody will reprint.
Sources
[1] Benj Edwards, “Anthropic destroyed millions of print books to build its AI models,” Ars Technica, 25 June 2025. arstechnica.com/ai/2025/06/anthropic-destroyed-millions-of-print-books-to-build-its-ai-models/
[2] “Federal Judge Rules AI Training Is Fair Use in Anthropic Copyright Case,” Publishers Weekly, 25 June 2025 — Bartz v. Anthropic, N.D. Cal. publishersweekly.com/pw/by-topic/digital/copyright/article/98089-federal-judge-rules-ai-training-is-fair-use-in-anthropic-copyright-case.html
[3] “Mixed Decision in Anthropic AI Case,” The Authors Guild. authorsguild.org/news/mixed-decision-in-anthropic-ai-case/
[4] Associated Press, “Anthropic pays $1.5B to authors in Claude AI copyright settlement over pirated books” — final approval 20 July 2026. aol.com/articles/anthropic-pays-1-5b-authors-170300000.html
[5] Samantha Cole, “AI Companies Are Buying Tons of Old Books Because They’re Free of AI Slop,” 404 Media, 21 July 2026.
[6] “Are AI companies scanning and destroying millions of books, including rare titles?” Snopes. snopes.com/fact-check/ai-companies-destroying-rare-books/
[7] “Do Americans read print books, e-books or audiobooks more?” Pew Research Center — survey October 2025, published 9 April 2026. pewresearch.org/short-reads/2026/04/09/americans-still-opt-for-print-books-over-digital-or-audio-versions-few-are-in-book-clubs/